1 //===- SourceManager.cpp - Track and cache source files -------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file implements the SourceManager interface. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "clang/Basic/SourceManager.h" 14 #include "clang/Basic/Diagnostic.h" 15 #include "clang/Basic/FileManager.h" 16 #include "clang/Basic/LLVM.h" 17 #include "clang/Basic/SourceLocation.h" 18 #include "clang/Basic/SourceManagerInternals.h" 19 #include "llvm/ADT/DenseMap.h" 20 #include "llvm/ADT/Optional.h" 21 #include "llvm/ADT/None.h" 22 #include "llvm/ADT/STLExtras.h" 23 #include "llvm/ADT/SmallVector.h" 24 #include "llvm/ADT/StringSwitch.h" 25 #include "llvm/ADT/StringRef.h" 26 #include "llvm/Support/Allocator.h" 27 #include "llvm/Support/Capacity.h" 28 #include "llvm/Support/Compiler.h" 29 #include "llvm/Support/ErrorHandling.h" 30 #include "llvm/Support/FileSystem.h" 31 #include "llvm/Support/MathExtras.h" 32 #include "llvm/Support/MemoryBuffer.h" 33 #include "llvm/Support/Path.h" 34 #include "llvm/Support/raw_ostream.h" 35 #include <algorithm> 36 #include <cassert> 37 #include <cstddef> 38 #include <cstdint> 39 #include <memory> 40 #include <tuple> 41 #include <utility> 42 #include <vector> 43 44 using namespace clang; 45 using namespace SrcMgr; 46 using llvm::MemoryBuffer; 47 48 //===----------------------------------------------------------------------===// 49 // SourceManager Helper Classes 50 //===----------------------------------------------------------------------===// 51 52 ContentCache::~ContentCache() { 53 if (shouldFreeBuffer()) 54 delete Buffer.getPointer(); 55 } 56 57 /// getSizeBytesMapped - Returns the number of bytes actually mapped for this 58 /// ContentCache. This can be 0 if the MemBuffer was not actually expanded. 59 unsigned ContentCache::getSizeBytesMapped() const { 60 return Buffer.getPointer() ? Buffer.getPointer()->getBufferSize() : 0; 61 } 62 63 /// Returns the kind of memory used to back the memory buffer for 64 /// this content cache. This is used for performance analysis. 65 llvm::MemoryBuffer::BufferKind ContentCache::getMemoryBufferKind() const { 66 assert(Buffer.getPointer()); 67 68 // Should be unreachable, but keep for sanity. 69 if (!Buffer.getPointer()) 70 return llvm::MemoryBuffer::MemoryBuffer_Malloc; 71 72 const llvm::MemoryBuffer *buf = Buffer.getPointer(); 73 return buf->getBufferKind(); 74 } 75 76 /// getSize - Returns the size of the content encapsulated by this ContentCache. 77 /// This can be the size of the source file or the size of an arbitrary 78 /// scratch buffer. If the ContentCache encapsulates a source file, that 79 /// file is not lazily brought in from disk to satisfy this query. 80 unsigned ContentCache::getSize() const { 81 return Buffer.getPointer() ? (unsigned) Buffer.getPointer()->getBufferSize() 82 : (unsigned) ContentsEntry->getSize(); 83 } 84 85 void ContentCache::replaceBuffer(const llvm::MemoryBuffer *B, bool DoNotFree) { 86 if (B && B == Buffer.getPointer()) { 87 assert(0 && "Replacing with the same buffer"); 88 Buffer.setInt(DoNotFree? DoNotFreeFlag : 0); 89 return; 90 } 91 92 if (shouldFreeBuffer()) 93 delete Buffer.getPointer(); 94 Buffer.setPointer(B); 95 Buffer.setInt((B && DoNotFree) ? DoNotFreeFlag : 0); 96 } 97 98 const llvm::MemoryBuffer *ContentCache::getBuffer(DiagnosticsEngine &Diag, 99 FileManager &FM, 100 SourceLocation Loc, 101 bool *Invalid) const { 102 // Lazily create the Buffer for ContentCaches that wrap files. If we already 103 // computed it, just return what we have. 104 if (Buffer.getPointer() || !ContentsEntry) { 105 if (Invalid) 106 *Invalid = isBufferInvalid(); 107 108 return Buffer.getPointer(); 109 } 110 111 // Check that the file's size fits in an 'unsigned' (with room for a 112 // past-the-end value). This is deeply regrettable, but various parts of 113 // Clang (including elsewhere in this file!) use 'unsigned' to represent file 114 // offsets, line numbers, string literal lengths, and so on, and fail 115 // miserably on large source files. 116 if ((uint64_t)ContentsEntry->getSize() >= 117 std::numeric_limits<unsigned>::max()) { 118 // We can't make a memory buffer of the required size, so just make a small 119 // one. We should never hit a situation where we've already parsed to a 120 // later offset of the file, so it shouldn't matter that the buffer is 121 // smaller than the file. 122 Buffer.setPointer( 123 llvm::MemoryBuffer::getMemBuffer("", ContentsEntry->getName()) 124 .release()); 125 if (Diag.isDiagnosticInFlight()) 126 Diag.SetDelayedDiagnostic(diag::err_file_too_large, 127 ContentsEntry->getName()); 128 else 129 Diag.Report(Loc, diag::err_file_too_large) 130 << ContentsEntry->getName(); 131 132 Buffer.setInt(Buffer.getInt() | InvalidFlag); 133 if (Invalid) *Invalid = true; 134 return Buffer.getPointer(); 135 } 136 137 auto BufferOrError = FM.getBufferForFile(ContentsEntry, IsFileVolatile); 138 139 // If we were unable to open the file, then we are in an inconsistent 140 // situation where the content cache referenced a file which no longer 141 // exists. Most likely, we were using a stat cache with an invalid entry but 142 // the file could also have been removed during processing. Since we can't 143 // really deal with this situation, just create an empty buffer. 144 // 145 // FIXME: This is definitely not ideal, but our immediate clients can't 146 // currently handle returning a null entry here. Ideally we should detect 147 // that we are in an inconsistent situation and error out as quickly as 148 // possible. 149 if (!BufferOrError) { 150 StringRef FillStr("<<<MISSING SOURCE FILE>>>\n"); 151 auto BackupBuffer = llvm::WritableMemoryBuffer::getNewUninitMemBuffer( 152 ContentsEntry->getSize(), "<invalid>"); 153 char *Ptr = BackupBuffer->getBufferStart(); 154 for (unsigned i = 0, e = ContentsEntry->getSize(); i != e; ++i) 155 Ptr[i] = FillStr[i % FillStr.size()]; 156 Buffer.setPointer(BackupBuffer.release()); 157 158 if (Diag.isDiagnosticInFlight()) 159 Diag.SetDelayedDiagnostic(diag::err_cannot_open_file, 160 ContentsEntry->getName(), 161 BufferOrError.getError().message()); 162 else 163 Diag.Report(Loc, diag::err_cannot_open_file) 164 << ContentsEntry->getName() << BufferOrError.getError().message(); 165 166 Buffer.setInt(Buffer.getInt() | InvalidFlag); 167 168 if (Invalid) *Invalid = true; 169 return Buffer.getPointer(); 170 } 171 172 Buffer.setPointer(BufferOrError->release()); 173 174 // Check that the file's size is the same as in the file entry (which may 175 // have come from a stat cache). 176 if (getRawBuffer()->getBufferSize() != (size_t)ContentsEntry->getSize()) { 177 if (Diag.isDiagnosticInFlight()) 178 Diag.SetDelayedDiagnostic(diag::err_file_modified, 179 ContentsEntry->getName()); 180 else 181 Diag.Report(Loc, diag::err_file_modified) 182 << ContentsEntry->getName(); 183 184 Buffer.setInt(Buffer.getInt() | InvalidFlag); 185 if (Invalid) *Invalid = true; 186 return Buffer.getPointer(); 187 } 188 189 // If the buffer is valid, check to see if it has a UTF Byte Order Mark 190 // (BOM). We only support UTF-8 with and without a BOM right now. See 191 // http://en.wikipedia.org/wiki/Byte_order_mark for more information. 192 StringRef BufStr = Buffer.getPointer()->getBuffer(); 193 const char *InvalidBOM = llvm::StringSwitch<const char *>(BufStr) 194 .StartsWith(llvm::StringLiteral::withInnerNUL("\x00\x00\xFE\xFF"), 195 "UTF-32 (BE)") 196 .StartsWith(llvm::StringLiteral::withInnerNUL("\xFF\xFE\x00\x00"), 197 "UTF-32 (LE)") 198 .StartsWith("\xFE\xFF", "UTF-16 (BE)") 199 .StartsWith("\xFF\xFE", "UTF-16 (LE)") 200 .StartsWith("\x2B\x2F\x76", "UTF-7") 201 .StartsWith("\xF7\x64\x4C", "UTF-1") 202 .StartsWith("\xDD\x73\x66\x73", "UTF-EBCDIC") 203 .StartsWith("\x0E\xFE\xFF", "SCSU") 204 .StartsWith("\xFB\xEE\x28", "BOCU-1") 205 .StartsWith("\x84\x31\x95\x33", "GB-18030") 206 .Default(nullptr); 207 208 if (InvalidBOM) { 209 Diag.Report(Loc, diag::err_unsupported_bom) 210 << InvalidBOM << ContentsEntry->getName(); 211 Buffer.setInt(Buffer.getInt() | InvalidFlag); 212 } 213 214 if (Invalid) 215 *Invalid = isBufferInvalid(); 216 217 return Buffer.getPointer(); 218 } 219 220 unsigned LineTableInfo::getLineTableFilenameID(StringRef Name) { 221 auto IterBool = FilenameIDs.try_emplace(Name, FilenamesByID.size()); 222 if (IterBool.second) 223 FilenamesByID.push_back(&*IterBool.first); 224 return IterBool.first->second; 225 } 226 227 /// Add a line note to the line table that indicates that there is a \#line or 228 /// GNU line marker at the specified FID/Offset location which changes the 229 /// presumed location to LineNo/FilenameID. If EntryExit is 0, then this doesn't 230 /// change the presumed \#include stack. If it is 1, this is a file entry, if 231 /// it is 2 then this is a file exit. FileKind specifies whether this is a 232 /// system header or extern C system header. 233 void LineTableInfo::AddLineNote(FileID FID, unsigned Offset, unsigned LineNo, 234 int FilenameID, unsigned EntryExit, 235 SrcMgr::CharacteristicKind FileKind) { 236 std::vector<LineEntry> &Entries = LineEntries[FID]; 237 238 // An unspecified FilenameID means use the last filename if available, or the 239 // main source file otherwise. 240 if (FilenameID == -1 && !Entries.empty()) 241 FilenameID = Entries.back().FilenameID; 242 243 assert((Entries.empty() || Entries.back().FileOffset < Offset) && 244 "Adding line entries out of order!"); 245 246 unsigned IncludeOffset = 0; 247 if (EntryExit == 0) { // No #include stack change. 248 IncludeOffset = Entries.empty() ? 0 : Entries.back().IncludeOffset; 249 } else if (EntryExit == 1) { 250 IncludeOffset = Offset-1; 251 } else if (EntryExit == 2) { 252 assert(!Entries.empty() && Entries.back().IncludeOffset && 253 "PPDirectives should have caught case when popping empty include stack"); 254 255 // Get the include loc of the last entries' include loc as our include loc. 256 IncludeOffset = 0; 257 if (const LineEntry *PrevEntry = 258 FindNearestLineEntry(FID, Entries.back().IncludeOffset)) 259 IncludeOffset = PrevEntry->IncludeOffset; 260 } 261 262 Entries.push_back(LineEntry::get(Offset, LineNo, FilenameID, FileKind, 263 IncludeOffset)); 264 } 265 266 /// FindNearestLineEntry - Find the line entry nearest to FID that is before 267 /// it. If there is no line entry before Offset in FID, return null. 268 const LineEntry *LineTableInfo::FindNearestLineEntry(FileID FID, 269 unsigned Offset) { 270 const std::vector<LineEntry> &Entries = LineEntries[FID]; 271 assert(!Entries.empty() && "No #line entries for this FID after all!"); 272 273 // It is very common for the query to be after the last #line, check this 274 // first. 275 if (Entries.back().FileOffset <= Offset) 276 return &Entries.back(); 277 278 // Do a binary search to find the maximal element that is still before Offset. 279 std::vector<LineEntry>::const_iterator I = llvm::upper_bound(Entries, Offset); 280 if (I == Entries.begin()) 281 return nullptr; 282 return &*--I; 283 } 284 285 /// Add a new line entry that has already been encoded into 286 /// the internal representation of the line table. 287 void LineTableInfo::AddEntry(FileID FID, 288 const std::vector<LineEntry> &Entries) { 289 LineEntries[FID] = Entries; 290 } 291 292 /// getLineTableFilenameID - Return the uniqued ID for the specified filename. 293 unsigned SourceManager::getLineTableFilenameID(StringRef Name) { 294 return getLineTable().getLineTableFilenameID(Name); 295 } 296 297 /// AddLineNote - Add a line note to the line table for the FileID and offset 298 /// specified by Loc. If FilenameID is -1, it is considered to be 299 /// unspecified. 300 void SourceManager::AddLineNote(SourceLocation Loc, unsigned LineNo, 301 int FilenameID, bool IsFileEntry, 302 bool IsFileExit, 303 SrcMgr::CharacteristicKind FileKind) { 304 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 305 306 bool Invalid = false; 307 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 308 if (!Entry.isFile() || Invalid) 309 return; 310 311 const SrcMgr::FileInfo &FileInfo = Entry.getFile(); 312 313 // Remember that this file has #line directives now if it doesn't already. 314 const_cast<SrcMgr::FileInfo&>(FileInfo).setHasLineDirectives(); 315 316 (void) getLineTable(); 317 318 unsigned EntryExit = 0; 319 if (IsFileEntry) 320 EntryExit = 1; 321 else if (IsFileExit) 322 EntryExit = 2; 323 324 LineTable->AddLineNote(LocInfo.first, LocInfo.second, LineNo, FilenameID, 325 EntryExit, FileKind); 326 } 327 328 LineTableInfo &SourceManager::getLineTable() { 329 if (!LineTable) 330 LineTable.reset(new LineTableInfo()); 331 return *LineTable; 332 } 333 334 //===----------------------------------------------------------------------===// 335 // Private 'Create' methods. 336 //===----------------------------------------------------------------------===// 337 338 SourceManager::SourceManager(DiagnosticsEngine &Diag, FileManager &FileMgr, 339 bool UserFilesAreVolatile) 340 : Diag(Diag), FileMgr(FileMgr), UserFilesAreVolatile(UserFilesAreVolatile) { 341 clearIDTables(); 342 Diag.setSourceManager(this); 343 } 344 345 SourceManager::~SourceManager() { 346 // Delete FileEntry objects corresponding to content caches. Since the actual 347 // content cache objects are bump pointer allocated, we just have to run the 348 // dtors, but we call the deallocate method for completeness. 349 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) { 350 if (MemBufferInfos[i]) { 351 MemBufferInfos[i]->~ContentCache(); 352 ContentCacheAlloc.Deallocate(MemBufferInfos[i]); 353 } 354 } 355 for (llvm::DenseMap<const FileEntry*, SrcMgr::ContentCache*>::iterator 356 I = FileInfos.begin(), E = FileInfos.end(); I != E; ++I) { 357 if (I->second) { 358 I->second->~ContentCache(); 359 ContentCacheAlloc.Deallocate(I->second); 360 } 361 } 362 } 363 364 void SourceManager::clearIDTables() { 365 MainFileID = FileID(); 366 LocalSLocEntryTable.clear(); 367 LoadedSLocEntryTable.clear(); 368 SLocEntryLoaded.clear(); 369 LastLineNoFileIDQuery = FileID(); 370 LastLineNoContentCache = nullptr; 371 LastFileIDLookup = FileID(); 372 373 if (LineTable) 374 LineTable->clear(); 375 376 // Use up FileID #0 as an invalid expansion. 377 NextLocalOffset = 0; 378 CurrentLoadedOffset = MaxLoadedOffset; 379 createExpansionLoc(SourceLocation(), SourceLocation(), SourceLocation(), 1); 380 } 381 382 void SourceManager::initializeForReplay(const SourceManager &Old) { 383 assert(MainFileID.isInvalid() && "expected uninitialized SourceManager"); 384 385 auto CloneContentCache = [&](const ContentCache *Cache) -> ContentCache * { 386 auto *Clone = new (ContentCacheAlloc.Allocate<ContentCache>()) ContentCache; 387 Clone->OrigEntry = Cache->OrigEntry; 388 Clone->ContentsEntry = Cache->ContentsEntry; 389 Clone->BufferOverridden = Cache->BufferOverridden; 390 Clone->IsFileVolatile = Cache->IsFileVolatile; 391 Clone->IsTransient = Cache->IsTransient; 392 Clone->replaceBuffer(Cache->getRawBuffer(), /*DoNotFree*/true); 393 return Clone; 394 }; 395 396 // Ensure all SLocEntries are loaded from the external source. 397 for (unsigned I = 0, N = Old.LoadedSLocEntryTable.size(); I != N; ++I) 398 if (!Old.SLocEntryLoaded[I]) 399 Old.loadSLocEntry(I, nullptr); 400 401 // Inherit any content cache data from the old source manager. 402 for (auto &FileInfo : Old.FileInfos) { 403 SrcMgr::ContentCache *&Slot = FileInfos[FileInfo.first]; 404 if (Slot) 405 continue; 406 Slot = CloneContentCache(FileInfo.second); 407 } 408 } 409 410 /// getOrCreateContentCache - Create or return a cached ContentCache for the 411 /// specified file. 412 const ContentCache * 413 SourceManager::getOrCreateContentCache(const FileEntry *FileEnt, 414 bool isSystemFile) { 415 assert(FileEnt && "Didn't specify a file entry to use?"); 416 417 // Do we already have information about this file? 418 ContentCache *&Entry = FileInfos[FileEnt]; 419 if (Entry) return Entry; 420 421 // Nope, create a new Cache entry. 422 Entry = ContentCacheAlloc.Allocate<ContentCache>(); 423 424 if (OverriddenFilesInfo) { 425 // If the file contents are overridden with contents from another file, 426 // pass that file to ContentCache. 427 llvm::DenseMap<const FileEntry *, const FileEntry *>::iterator 428 overI = OverriddenFilesInfo->OverriddenFiles.find(FileEnt); 429 if (overI == OverriddenFilesInfo->OverriddenFiles.end()) 430 new (Entry) ContentCache(FileEnt); 431 else 432 new (Entry) ContentCache(OverridenFilesKeepOriginalName ? FileEnt 433 : overI->second, 434 overI->second); 435 } else { 436 new (Entry) ContentCache(FileEnt); 437 } 438 439 Entry->IsFileVolatile = UserFilesAreVolatile && !isSystemFile; 440 Entry->IsTransient = FilesAreTransient; 441 442 return Entry; 443 } 444 445 /// Create a new ContentCache for the specified memory buffer. 446 /// This does no caching. 447 const ContentCache * 448 SourceManager::createMemBufferContentCache(const llvm::MemoryBuffer *Buffer, 449 bool DoNotFree) { 450 // Add a new ContentCache to the MemBufferInfos list and return it. 451 ContentCache *Entry = ContentCacheAlloc.Allocate<ContentCache>(); 452 new (Entry) ContentCache(); 453 MemBufferInfos.push_back(Entry); 454 Entry->replaceBuffer(Buffer, DoNotFree); 455 return Entry; 456 } 457 458 const SrcMgr::SLocEntry &SourceManager::loadSLocEntry(unsigned Index, 459 bool *Invalid) const { 460 assert(!SLocEntryLoaded[Index]); 461 if (ExternalSLocEntries->ReadSLocEntry(-(static_cast<int>(Index) + 2))) { 462 if (Invalid) 463 *Invalid = true; 464 // If the file of the SLocEntry changed we could still have loaded it. 465 if (!SLocEntryLoaded[Index]) { 466 // Try to recover; create a SLocEntry so the rest of clang can handle it. 467 LoadedSLocEntryTable[Index] = SLocEntry::get( 468 0, FileInfo::get(SourceLocation(), getFakeContentCacheForRecovery(), 469 SrcMgr::C_User, "")); 470 } 471 } 472 473 return LoadedSLocEntryTable[Index]; 474 } 475 476 std::pair<int, unsigned> 477 SourceManager::AllocateLoadedSLocEntries(unsigned NumSLocEntries, 478 unsigned TotalSize) { 479 assert(ExternalSLocEntries && "Don't have an external sloc source"); 480 // Make sure we're not about to run out of source locations. 481 if (CurrentLoadedOffset - TotalSize < NextLocalOffset) 482 return std::make_pair(0, 0); 483 LoadedSLocEntryTable.resize(LoadedSLocEntryTable.size() + NumSLocEntries); 484 SLocEntryLoaded.resize(LoadedSLocEntryTable.size()); 485 CurrentLoadedOffset -= TotalSize; 486 int ID = LoadedSLocEntryTable.size(); 487 return std::make_pair(-ID - 1, CurrentLoadedOffset); 488 } 489 490 /// As part of recovering from missing or changed content, produce a 491 /// fake, non-empty buffer. 492 llvm::MemoryBuffer *SourceManager::getFakeBufferForRecovery() const { 493 if (!FakeBufferForRecovery) 494 FakeBufferForRecovery = 495 llvm::MemoryBuffer::getMemBuffer("<<<INVALID BUFFER>>"); 496 497 return FakeBufferForRecovery.get(); 498 } 499 500 /// As part of recovering from missing or changed content, produce a 501 /// fake content cache. 502 const SrcMgr::ContentCache * 503 SourceManager::getFakeContentCacheForRecovery() const { 504 if (!FakeContentCacheForRecovery) { 505 FakeContentCacheForRecovery = std::make_unique<SrcMgr::ContentCache>(); 506 FakeContentCacheForRecovery->replaceBuffer(getFakeBufferForRecovery(), 507 /*DoNotFree=*/true); 508 } 509 return FakeContentCacheForRecovery.get(); 510 } 511 512 /// Returns the previous in-order FileID or an invalid FileID if there 513 /// is no previous one. 514 FileID SourceManager::getPreviousFileID(FileID FID) const { 515 if (FID.isInvalid()) 516 return FileID(); 517 518 int ID = FID.ID; 519 if (ID == -1) 520 return FileID(); 521 522 if (ID > 0) { 523 if (ID-1 == 0) 524 return FileID(); 525 } else if (unsigned(-(ID-1) - 2) >= LoadedSLocEntryTable.size()) { 526 return FileID(); 527 } 528 529 return FileID::get(ID-1); 530 } 531 532 /// Returns the next in-order FileID or an invalid FileID if there is 533 /// no next one. 534 FileID SourceManager::getNextFileID(FileID FID) const { 535 if (FID.isInvalid()) 536 return FileID(); 537 538 int ID = FID.ID; 539 if (ID > 0) { 540 if (unsigned(ID+1) >= local_sloc_entry_size()) 541 return FileID(); 542 } else if (ID+1 >= -1) { 543 return FileID(); 544 } 545 546 return FileID::get(ID+1); 547 } 548 549 //===----------------------------------------------------------------------===// 550 // Methods to create new FileID's and macro expansions. 551 //===----------------------------------------------------------------------===// 552 553 /// createFileID - Create a new FileID for the specified ContentCache and 554 /// include position. This works regardless of whether the ContentCache 555 /// corresponds to a file or some other input source. 556 FileID SourceManager::createFileID(const ContentCache *File, StringRef Filename, 557 SourceLocation IncludePos, 558 SrcMgr::CharacteristicKind FileCharacter, 559 int LoadedID, unsigned LoadedOffset) { 560 if (LoadedID < 0) { 561 assert(LoadedID != -1 && "Loading sentinel FileID"); 562 unsigned Index = unsigned(-LoadedID) - 2; 563 assert(Index < LoadedSLocEntryTable.size() && "FileID out of range"); 564 assert(!SLocEntryLoaded[Index] && "FileID already loaded"); 565 LoadedSLocEntryTable[Index] = SLocEntry::get( 566 LoadedOffset, FileInfo::get(IncludePos, File, FileCharacter, Filename)); 567 SLocEntryLoaded[Index] = true; 568 return FileID::get(LoadedID); 569 } 570 LocalSLocEntryTable.push_back( 571 SLocEntry::get(NextLocalOffset, 572 FileInfo::get(IncludePos, File, FileCharacter, Filename))); 573 unsigned FileSize = File->getSize(); 574 assert(NextLocalOffset + FileSize + 1 > NextLocalOffset && 575 NextLocalOffset + FileSize + 1 <= CurrentLoadedOffset && 576 "Ran out of source locations!"); 577 // We do a +1 here because we want a SourceLocation that means "the end of the 578 // file", e.g. for the "no newline at the end of the file" diagnostic. 579 NextLocalOffset += FileSize + 1; 580 581 // Set LastFileIDLookup to the newly created file. The next getFileID call is 582 // almost guaranteed to be from that file. 583 FileID FID = FileID::get(LocalSLocEntryTable.size()-1); 584 return LastFileIDLookup = FID; 585 } 586 587 SourceLocation 588 SourceManager::createMacroArgExpansionLoc(SourceLocation SpellingLoc, 589 SourceLocation ExpansionLoc, 590 unsigned TokLength) { 591 ExpansionInfo Info = ExpansionInfo::createForMacroArg(SpellingLoc, 592 ExpansionLoc); 593 return createExpansionLocImpl(Info, TokLength); 594 } 595 596 SourceLocation 597 SourceManager::createExpansionLoc(SourceLocation SpellingLoc, 598 SourceLocation ExpansionLocStart, 599 SourceLocation ExpansionLocEnd, 600 unsigned TokLength, 601 bool ExpansionIsTokenRange, 602 int LoadedID, 603 unsigned LoadedOffset) { 604 ExpansionInfo Info = ExpansionInfo::create( 605 SpellingLoc, ExpansionLocStart, ExpansionLocEnd, ExpansionIsTokenRange); 606 return createExpansionLocImpl(Info, TokLength, LoadedID, LoadedOffset); 607 } 608 609 SourceLocation SourceManager::createTokenSplitLoc(SourceLocation Spelling, 610 SourceLocation TokenStart, 611 SourceLocation TokenEnd) { 612 assert(getFileID(TokenStart) == getFileID(TokenEnd) && 613 "token spans multiple files"); 614 return createExpansionLocImpl( 615 ExpansionInfo::createForTokenSplit(Spelling, TokenStart, TokenEnd), 616 TokenEnd.getOffset() - TokenStart.getOffset()); 617 } 618 619 SourceLocation 620 SourceManager::createExpansionLocImpl(const ExpansionInfo &Info, 621 unsigned TokLength, 622 int LoadedID, 623 unsigned LoadedOffset) { 624 if (LoadedID < 0) { 625 assert(LoadedID != -1 && "Loading sentinel FileID"); 626 unsigned Index = unsigned(-LoadedID) - 2; 627 assert(Index < LoadedSLocEntryTable.size() && "FileID out of range"); 628 assert(!SLocEntryLoaded[Index] && "FileID already loaded"); 629 LoadedSLocEntryTable[Index] = SLocEntry::get(LoadedOffset, Info); 630 SLocEntryLoaded[Index] = true; 631 return SourceLocation::getMacroLoc(LoadedOffset); 632 } 633 LocalSLocEntryTable.push_back(SLocEntry::get(NextLocalOffset, Info)); 634 assert(NextLocalOffset + TokLength + 1 > NextLocalOffset && 635 NextLocalOffset + TokLength + 1 <= CurrentLoadedOffset && 636 "Ran out of source locations!"); 637 // See createFileID for that +1. 638 NextLocalOffset += TokLength + 1; 639 return SourceLocation::getMacroLoc(NextLocalOffset - (TokLength + 1)); 640 } 641 642 const llvm::MemoryBuffer * 643 SourceManager::getMemoryBufferForFile(const FileEntry *File, bool *Invalid) { 644 const SrcMgr::ContentCache *IR = getOrCreateContentCache(File); 645 assert(IR && "getOrCreateContentCache() cannot return NULL"); 646 return IR->getBuffer(Diag, getFileManager(), SourceLocation(), Invalid); 647 } 648 649 void SourceManager::overrideFileContents(const FileEntry *SourceFile, 650 llvm::MemoryBuffer *Buffer, 651 bool DoNotFree) { 652 const SrcMgr::ContentCache *IR = getOrCreateContentCache(SourceFile); 653 assert(IR && "getOrCreateContentCache() cannot return NULL"); 654 655 const_cast<SrcMgr::ContentCache *>(IR)->replaceBuffer(Buffer, DoNotFree); 656 const_cast<SrcMgr::ContentCache *>(IR)->BufferOverridden = true; 657 658 getOverriddenFilesInfo().OverriddenFilesWithBuffer.insert(SourceFile); 659 } 660 661 void SourceManager::overrideFileContents(const FileEntry *SourceFile, 662 const FileEntry *NewFile) { 663 assert(SourceFile->getSize() == NewFile->getSize() && 664 "Different sizes, use the FileManager to create a virtual file with " 665 "the correct size"); 666 assert(FileInfos.count(SourceFile) == 0 && 667 "This function should be called at the initialization stage, before " 668 "any parsing occurs."); 669 getOverriddenFilesInfo().OverriddenFiles[SourceFile] = NewFile; 670 } 671 672 const FileEntry * 673 SourceManager::bypassFileContentsOverride(const FileEntry &File) { 674 assert(isFileOverridden(&File)); 675 llvm::Optional<FileEntryRef> BypassFile = 676 FileMgr.getBypassFile(FileEntryRef(File.getName(), File)); 677 678 // If the file can't be found in the FS, give up. 679 if (!BypassFile) 680 return nullptr; 681 682 const FileEntry *FE = &BypassFile->getFileEntry(); 683 (void)getOrCreateContentCache(FE); 684 return FE; 685 } 686 687 void SourceManager::setFileIsTransient(const FileEntry *File) { 688 const SrcMgr::ContentCache *CC = getOrCreateContentCache(File); 689 const_cast<SrcMgr::ContentCache *>(CC)->IsTransient = true; 690 } 691 692 StringRef SourceManager::getBufferData(FileID FID, bool *Invalid) const { 693 bool MyInvalid = false; 694 const SLocEntry &SLoc = getSLocEntry(FID, &MyInvalid); 695 if (!SLoc.isFile() || MyInvalid) { 696 if (Invalid) 697 *Invalid = true; 698 return "<<<<<INVALID SOURCE LOCATION>>>>>"; 699 } 700 701 const llvm::MemoryBuffer *Buf = SLoc.getFile().getContentCache()->getBuffer( 702 Diag, getFileManager(), SourceLocation(), &MyInvalid); 703 if (Invalid) 704 *Invalid = MyInvalid; 705 706 if (MyInvalid) 707 return "<<<<<INVALID SOURCE LOCATION>>>>>"; 708 709 return Buf->getBuffer(); 710 } 711 712 //===----------------------------------------------------------------------===// 713 // SourceLocation manipulation methods. 714 //===----------------------------------------------------------------------===// 715 716 /// Return the FileID for a SourceLocation. 717 /// 718 /// This is the cache-miss path of getFileID. Not as hot as that function, but 719 /// still very important. It is responsible for finding the entry in the 720 /// SLocEntry tables that contains the specified location. 721 FileID SourceManager::getFileIDSlow(unsigned SLocOffset) const { 722 if (!SLocOffset) 723 return FileID::get(0); 724 725 // Now it is time to search for the correct file. See where the SLocOffset 726 // sits in the global view and consult local or loaded buffers for it. 727 if (SLocOffset < NextLocalOffset) 728 return getFileIDLocal(SLocOffset); 729 return getFileIDLoaded(SLocOffset); 730 } 731 732 /// Return the FileID for a SourceLocation with a low offset. 733 /// 734 /// This function knows that the SourceLocation is in a local buffer, not a 735 /// loaded one. 736 FileID SourceManager::getFileIDLocal(unsigned SLocOffset) const { 737 assert(SLocOffset < NextLocalOffset && "Bad function choice"); 738 739 // After the first and second level caches, I see two common sorts of 740 // behavior: 1) a lot of searched FileID's are "near" the cached file 741 // location or are "near" the cached expansion location. 2) others are just 742 // completely random and may be a very long way away. 743 // 744 // To handle this, we do a linear search for up to 8 steps to catch #1 quickly 745 // then we fall back to a less cache efficient, but more scalable, binary 746 // search to find the location. 747 748 // See if this is near the file point - worst case we start scanning from the 749 // most newly created FileID. 750 const SrcMgr::SLocEntry *I; 751 752 if (LastFileIDLookup.ID < 0 || 753 LocalSLocEntryTable[LastFileIDLookup.ID].getOffset() < SLocOffset) { 754 // Neither loc prunes our search. 755 I = LocalSLocEntryTable.end(); 756 } else { 757 // Perhaps it is near the file point. 758 I = LocalSLocEntryTable.begin()+LastFileIDLookup.ID; 759 } 760 761 // Find the FileID that contains this. "I" is an iterator that points to a 762 // FileID whose offset is known to be larger than SLocOffset. 763 unsigned NumProbes = 0; 764 while (true) { 765 --I; 766 if (I->getOffset() <= SLocOffset) { 767 FileID Res = FileID::get(int(I - LocalSLocEntryTable.begin())); 768 769 // If this isn't an expansion, remember it. We have good locality across 770 // FileID lookups. 771 if (!I->isExpansion()) 772 LastFileIDLookup = Res; 773 NumLinearScans += NumProbes+1; 774 return Res; 775 } 776 if (++NumProbes == 8) 777 break; 778 } 779 780 // Convert "I" back into an index. We know that it is an entry whose index is 781 // larger than the offset we are looking for. 782 unsigned GreaterIndex = I - LocalSLocEntryTable.begin(); 783 // LessIndex - This is the lower bound of the range that we're searching. 784 // We know that the offset corresponding to the FileID is is less than 785 // SLocOffset. 786 unsigned LessIndex = 0; 787 NumProbes = 0; 788 while (true) { 789 bool Invalid = false; 790 unsigned MiddleIndex = (GreaterIndex-LessIndex)/2+LessIndex; 791 unsigned MidOffset = getLocalSLocEntry(MiddleIndex, &Invalid).getOffset(); 792 if (Invalid) 793 return FileID::get(0); 794 795 ++NumProbes; 796 797 // If the offset of the midpoint is too large, chop the high side of the 798 // range to the midpoint. 799 if (MidOffset > SLocOffset) { 800 GreaterIndex = MiddleIndex; 801 continue; 802 } 803 804 // If the middle index contains the value, succeed and return. 805 // FIXME: This could be made faster by using a function that's aware of 806 // being in the local area. 807 if (isOffsetInFileID(FileID::get(MiddleIndex), SLocOffset)) { 808 FileID Res = FileID::get(MiddleIndex); 809 810 // If this isn't a macro expansion, remember it. We have good locality 811 // across FileID lookups. 812 if (!LocalSLocEntryTable[MiddleIndex].isExpansion()) 813 LastFileIDLookup = Res; 814 NumBinaryProbes += NumProbes; 815 return Res; 816 } 817 818 // Otherwise, move the low-side up to the middle index. 819 LessIndex = MiddleIndex; 820 } 821 } 822 823 /// Return the FileID for a SourceLocation with a high offset. 824 /// 825 /// This function knows that the SourceLocation is in a loaded buffer, not a 826 /// local one. 827 FileID SourceManager::getFileIDLoaded(unsigned SLocOffset) const { 828 // Sanity checking, otherwise a bug may lead to hanging in release build. 829 if (SLocOffset < CurrentLoadedOffset) { 830 assert(0 && "Invalid SLocOffset or bad function choice"); 831 return FileID(); 832 } 833 834 // Essentially the same as the local case, but the loaded array is sorted 835 // in the other direction. 836 837 // First do a linear scan from the last lookup position, if possible. 838 unsigned I; 839 int LastID = LastFileIDLookup.ID; 840 if (LastID >= 0 || getLoadedSLocEntryByID(LastID).getOffset() < SLocOffset) 841 I = 0; 842 else 843 I = (-LastID - 2) + 1; 844 845 unsigned NumProbes; 846 for (NumProbes = 0; NumProbes < 8; ++NumProbes, ++I) { 847 // Make sure the entry is loaded! 848 const SrcMgr::SLocEntry &E = getLoadedSLocEntry(I); 849 if (E.getOffset() <= SLocOffset) { 850 FileID Res = FileID::get(-int(I) - 2); 851 852 if (!E.isExpansion()) 853 LastFileIDLookup = Res; 854 NumLinearScans += NumProbes + 1; 855 return Res; 856 } 857 } 858 859 // Linear scan failed. Do the binary search. Note the reverse sorting of the 860 // table: GreaterIndex is the one where the offset is greater, which is 861 // actually a lower index! 862 unsigned GreaterIndex = I; 863 unsigned LessIndex = LoadedSLocEntryTable.size(); 864 NumProbes = 0; 865 while (true) { 866 ++NumProbes; 867 unsigned MiddleIndex = (LessIndex - GreaterIndex) / 2 + GreaterIndex; 868 const SrcMgr::SLocEntry &E = getLoadedSLocEntry(MiddleIndex); 869 if (E.getOffset() == 0) 870 return FileID(); // invalid entry. 871 872 ++NumProbes; 873 874 if (E.getOffset() > SLocOffset) { 875 // Sanity checking, otherwise a bug may lead to hanging in release build. 876 if (GreaterIndex == MiddleIndex) { 877 assert(0 && "binary search missed the entry"); 878 return FileID(); 879 } 880 GreaterIndex = MiddleIndex; 881 continue; 882 } 883 884 if (isOffsetInFileID(FileID::get(-int(MiddleIndex) - 2), SLocOffset)) { 885 FileID Res = FileID::get(-int(MiddleIndex) - 2); 886 if (!E.isExpansion()) 887 LastFileIDLookup = Res; 888 NumBinaryProbes += NumProbes; 889 return Res; 890 } 891 892 // Sanity checking, otherwise a bug may lead to hanging in release build. 893 if (LessIndex == MiddleIndex) { 894 assert(0 && "binary search missed the entry"); 895 return FileID(); 896 } 897 LessIndex = MiddleIndex; 898 } 899 } 900 901 SourceLocation SourceManager:: 902 getExpansionLocSlowCase(SourceLocation Loc) const { 903 do { 904 // Note: If Loc indicates an offset into a token that came from a macro 905 // expansion (e.g. the 5th character of the token) we do not want to add 906 // this offset when going to the expansion location. The expansion 907 // location is the macro invocation, which the offset has nothing to do 908 // with. This is unlike when we get the spelling loc, because the offset 909 // directly correspond to the token whose spelling we're inspecting. 910 Loc = getSLocEntry(getFileID(Loc)).getExpansion().getExpansionLocStart(); 911 } while (!Loc.isFileID()); 912 913 return Loc; 914 } 915 916 SourceLocation SourceManager::getSpellingLocSlowCase(SourceLocation Loc) const { 917 do { 918 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 919 Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc(); 920 Loc = Loc.getLocWithOffset(LocInfo.second); 921 } while (!Loc.isFileID()); 922 return Loc; 923 } 924 925 SourceLocation SourceManager::getFileLocSlowCase(SourceLocation Loc) const { 926 do { 927 if (isMacroArgExpansion(Loc)) 928 Loc = getImmediateSpellingLoc(Loc); 929 else 930 Loc = getImmediateExpansionRange(Loc).getBegin(); 931 } while (!Loc.isFileID()); 932 return Loc; 933 } 934 935 936 std::pair<FileID, unsigned> 937 SourceManager::getDecomposedExpansionLocSlowCase( 938 const SrcMgr::SLocEntry *E) const { 939 // If this is an expansion record, walk through all the expansion points. 940 FileID FID; 941 SourceLocation Loc; 942 unsigned Offset; 943 do { 944 Loc = E->getExpansion().getExpansionLocStart(); 945 946 FID = getFileID(Loc); 947 E = &getSLocEntry(FID); 948 Offset = Loc.getOffset()-E->getOffset(); 949 } while (!Loc.isFileID()); 950 951 return std::make_pair(FID, Offset); 952 } 953 954 std::pair<FileID, unsigned> 955 SourceManager::getDecomposedSpellingLocSlowCase(const SrcMgr::SLocEntry *E, 956 unsigned Offset) const { 957 // If this is an expansion record, walk through all the expansion points. 958 FileID FID; 959 SourceLocation Loc; 960 do { 961 Loc = E->getExpansion().getSpellingLoc(); 962 Loc = Loc.getLocWithOffset(Offset); 963 964 FID = getFileID(Loc); 965 E = &getSLocEntry(FID); 966 Offset = Loc.getOffset()-E->getOffset(); 967 } while (!Loc.isFileID()); 968 969 return std::make_pair(FID, Offset); 970 } 971 972 /// getImmediateSpellingLoc - Given a SourceLocation object, return the 973 /// spelling location referenced by the ID. This is the first level down 974 /// towards the place where the characters that make up the lexed token can be 975 /// found. This should not generally be used by clients. 976 SourceLocation SourceManager::getImmediateSpellingLoc(SourceLocation Loc) const{ 977 if (Loc.isFileID()) return Loc; 978 std::pair<FileID, unsigned> LocInfo = getDecomposedLoc(Loc); 979 Loc = getSLocEntry(LocInfo.first).getExpansion().getSpellingLoc(); 980 return Loc.getLocWithOffset(LocInfo.second); 981 } 982 983 /// getImmediateExpansionRange - Loc is required to be an expansion location. 984 /// Return the start/end of the expansion information. 985 CharSourceRange 986 SourceManager::getImmediateExpansionRange(SourceLocation Loc) const { 987 assert(Loc.isMacroID() && "Not a macro expansion loc!"); 988 const ExpansionInfo &Expansion = getSLocEntry(getFileID(Loc)).getExpansion(); 989 return Expansion.getExpansionLocRange(); 990 } 991 992 SourceLocation SourceManager::getTopMacroCallerLoc(SourceLocation Loc) const { 993 while (isMacroArgExpansion(Loc)) 994 Loc = getImmediateSpellingLoc(Loc); 995 return Loc; 996 } 997 998 /// getExpansionRange - Given a SourceLocation object, return the range of 999 /// tokens covered by the expansion in the ultimate file. 1000 CharSourceRange SourceManager::getExpansionRange(SourceLocation Loc) const { 1001 if (Loc.isFileID()) 1002 return CharSourceRange(SourceRange(Loc, Loc), true); 1003 1004 CharSourceRange Res = getImmediateExpansionRange(Loc); 1005 1006 // Fully resolve the start and end locations to their ultimate expansion 1007 // points. 1008 while (!Res.getBegin().isFileID()) 1009 Res.setBegin(getImmediateExpansionRange(Res.getBegin()).getBegin()); 1010 while (!Res.getEnd().isFileID()) { 1011 CharSourceRange EndRange = getImmediateExpansionRange(Res.getEnd()); 1012 Res.setEnd(EndRange.getEnd()); 1013 Res.setTokenRange(EndRange.isTokenRange()); 1014 } 1015 return Res; 1016 } 1017 1018 bool SourceManager::isMacroArgExpansion(SourceLocation Loc, 1019 SourceLocation *StartLoc) const { 1020 if (!Loc.isMacroID()) return false; 1021 1022 FileID FID = getFileID(Loc); 1023 const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion(); 1024 if (!Expansion.isMacroArgExpansion()) return false; 1025 1026 if (StartLoc) 1027 *StartLoc = Expansion.getExpansionLocStart(); 1028 return true; 1029 } 1030 1031 bool SourceManager::isMacroBodyExpansion(SourceLocation Loc) const { 1032 if (!Loc.isMacroID()) return false; 1033 1034 FileID FID = getFileID(Loc); 1035 const SrcMgr::ExpansionInfo &Expansion = getSLocEntry(FID).getExpansion(); 1036 return Expansion.isMacroBodyExpansion(); 1037 } 1038 1039 bool SourceManager::isAtStartOfImmediateMacroExpansion(SourceLocation Loc, 1040 SourceLocation *MacroBegin) const { 1041 assert(Loc.isValid() && Loc.isMacroID() && "Expected a valid macro loc"); 1042 1043 std::pair<FileID, unsigned> DecompLoc = getDecomposedLoc(Loc); 1044 if (DecompLoc.second > 0) 1045 return false; // Does not point at the start of expansion range. 1046 1047 bool Invalid = false; 1048 const SrcMgr::ExpansionInfo &ExpInfo = 1049 getSLocEntry(DecompLoc.first, &Invalid).getExpansion(); 1050 if (Invalid) 1051 return false; 1052 SourceLocation ExpLoc = ExpInfo.getExpansionLocStart(); 1053 1054 if (ExpInfo.isMacroArgExpansion()) { 1055 // For macro argument expansions, check if the previous FileID is part of 1056 // the same argument expansion, in which case this Loc is not at the 1057 // beginning of the expansion. 1058 FileID PrevFID = getPreviousFileID(DecompLoc.first); 1059 if (!PrevFID.isInvalid()) { 1060 const SrcMgr::SLocEntry &PrevEntry = getSLocEntry(PrevFID, &Invalid); 1061 if (Invalid) 1062 return false; 1063 if (PrevEntry.isExpansion() && 1064 PrevEntry.getExpansion().getExpansionLocStart() == ExpLoc) 1065 return false; 1066 } 1067 } 1068 1069 if (MacroBegin) 1070 *MacroBegin = ExpLoc; 1071 return true; 1072 } 1073 1074 bool SourceManager::isAtEndOfImmediateMacroExpansion(SourceLocation Loc, 1075 SourceLocation *MacroEnd) const { 1076 assert(Loc.isValid() && Loc.isMacroID() && "Expected a valid macro loc"); 1077 1078 FileID FID = getFileID(Loc); 1079 SourceLocation NextLoc = Loc.getLocWithOffset(1); 1080 if (isInFileID(NextLoc, FID)) 1081 return false; // Does not point at the end of expansion range. 1082 1083 bool Invalid = false; 1084 const SrcMgr::ExpansionInfo &ExpInfo = 1085 getSLocEntry(FID, &Invalid).getExpansion(); 1086 if (Invalid) 1087 return false; 1088 1089 if (ExpInfo.isMacroArgExpansion()) { 1090 // For macro argument expansions, check if the next FileID is part of the 1091 // same argument expansion, in which case this Loc is not at the end of the 1092 // expansion. 1093 FileID NextFID = getNextFileID(FID); 1094 if (!NextFID.isInvalid()) { 1095 const SrcMgr::SLocEntry &NextEntry = getSLocEntry(NextFID, &Invalid); 1096 if (Invalid) 1097 return false; 1098 if (NextEntry.isExpansion() && 1099 NextEntry.getExpansion().getExpansionLocStart() == 1100 ExpInfo.getExpansionLocStart()) 1101 return false; 1102 } 1103 } 1104 1105 if (MacroEnd) 1106 *MacroEnd = ExpInfo.getExpansionLocEnd(); 1107 return true; 1108 } 1109 1110 //===----------------------------------------------------------------------===// 1111 // Queries about the code at a SourceLocation. 1112 //===----------------------------------------------------------------------===// 1113 1114 /// getCharacterData - Return a pointer to the start of the specified location 1115 /// in the appropriate MemoryBuffer. 1116 const char *SourceManager::getCharacterData(SourceLocation SL, 1117 bool *Invalid) const { 1118 // Note that this is a hot function in the getSpelling() path, which is 1119 // heavily used by -E mode. 1120 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(SL); 1121 1122 // Note that calling 'getBuffer()' may lazily page in a source file. 1123 bool CharDataInvalid = false; 1124 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &CharDataInvalid); 1125 if (CharDataInvalid || !Entry.isFile()) { 1126 if (Invalid) 1127 *Invalid = true; 1128 1129 return "<<<<INVALID BUFFER>>>>"; 1130 } 1131 const llvm::MemoryBuffer *Buffer = 1132 Entry.getFile().getContentCache()->getBuffer( 1133 Diag, getFileManager(), SourceLocation(), &CharDataInvalid); 1134 if (Invalid) 1135 *Invalid = CharDataInvalid; 1136 return Buffer->getBufferStart() + (CharDataInvalid? 0 : LocInfo.second); 1137 } 1138 1139 /// getColumnNumber - Return the column # for the specified file position. 1140 /// this is significantly cheaper to compute than the line number. 1141 unsigned SourceManager::getColumnNumber(FileID FID, unsigned FilePos, 1142 bool *Invalid) const { 1143 bool MyInvalid = false; 1144 const llvm::MemoryBuffer *MemBuf = getBuffer(FID, &MyInvalid); 1145 if (Invalid) 1146 *Invalid = MyInvalid; 1147 1148 if (MyInvalid) 1149 return 1; 1150 1151 // It is okay to request a position just past the end of the buffer. 1152 if (FilePos > MemBuf->getBufferSize()) { 1153 if (Invalid) 1154 *Invalid = true; 1155 return 1; 1156 } 1157 1158 const char *Buf = MemBuf->getBufferStart(); 1159 // See if we just calculated the line number for this FilePos and can use 1160 // that to lookup the start of the line instead of searching for it. 1161 if (LastLineNoFileIDQuery == FID && 1162 LastLineNoContentCache->SourceLineCache != nullptr && 1163 LastLineNoResult < LastLineNoContentCache->NumLines) { 1164 unsigned *SourceLineCache = LastLineNoContentCache->SourceLineCache; 1165 unsigned LineStart = SourceLineCache[LastLineNoResult - 1]; 1166 unsigned LineEnd = SourceLineCache[LastLineNoResult]; 1167 if (FilePos >= LineStart && FilePos < LineEnd) { 1168 // LineEnd is the LineStart of the next line. 1169 // A line ends with separator LF or CR+LF on Windows. 1170 // FilePos might point to the last separator, 1171 // but we need a column number at most 1 + the last column. 1172 if (FilePos + 1 == LineEnd && FilePos > LineStart) { 1173 if (Buf[FilePos - 1] == '\r' || Buf[FilePos - 1] == '\n') 1174 --FilePos; 1175 } 1176 return FilePos - LineStart + 1; 1177 } 1178 } 1179 1180 unsigned LineStart = FilePos; 1181 while (LineStart && Buf[LineStart-1] != '\n' && Buf[LineStart-1] != '\r') 1182 --LineStart; 1183 return FilePos-LineStart+1; 1184 } 1185 1186 // isInvalid - Return the result of calling loc.isInvalid(), and 1187 // if Invalid is not null, set its value to same. 1188 template<typename LocType> 1189 static bool isInvalid(LocType Loc, bool *Invalid) { 1190 bool MyInvalid = Loc.isInvalid(); 1191 if (Invalid) 1192 *Invalid = MyInvalid; 1193 return MyInvalid; 1194 } 1195 1196 unsigned SourceManager::getSpellingColumnNumber(SourceLocation Loc, 1197 bool *Invalid) const { 1198 if (isInvalid(Loc, Invalid)) return 0; 1199 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 1200 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid); 1201 } 1202 1203 unsigned SourceManager::getExpansionColumnNumber(SourceLocation Loc, 1204 bool *Invalid) const { 1205 if (isInvalid(Loc, Invalid)) return 0; 1206 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1207 return getColumnNumber(LocInfo.first, LocInfo.second, Invalid); 1208 } 1209 1210 unsigned SourceManager::getPresumedColumnNumber(SourceLocation Loc, 1211 bool *Invalid) const { 1212 PresumedLoc PLoc = getPresumedLoc(Loc); 1213 if (isInvalid(PLoc, Invalid)) return 0; 1214 return PLoc.getColumn(); 1215 } 1216 1217 #ifdef __SSE2__ 1218 #include <emmintrin.h> 1219 #endif 1220 1221 static LLVM_ATTRIBUTE_NOINLINE void 1222 ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI, 1223 llvm::BumpPtrAllocator &Alloc, 1224 const SourceManager &SM, bool &Invalid); 1225 static void ComputeLineNumbers(DiagnosticsEngine &Diag, ContentCache *FI, 1226 llvm::BumpPtrAllocator &Alloc, 1227 const SourceManager &SM, bool &Invalid) { 1228 // Note that calling 'getBuffer()' may lazily page in the file. 1229 const MemoryBuffer *Buffer = 1230 FI->getBuffer(Diag, SM.getFileManager(), SourceLocation(), &Invalid); 1231 if (Invalid) 1232 return; 1233 1234 // Find the file offsets of all of the *physical* source lines. This does 1235 // not look at trigraphs, escaped newlines, or anything else tricky. 1236 SmallVector<unsigned, 256> LineOffsets; 1237 1238 // Line #1 starts at char 0. 1239 LineOffsets.push_back(0); 1240 1241 const unsigned char *Buf = (const unsigned char *)Buffer->getBufferStart(); 1242 const unsigned char *End = (const unsigned char *)Buffer->getBufferEnd(); 1243 unsigned I = 0; 1244 while (true) { 1245 // Skip over the contents of the line. 1246 while (Buf[I] != '\n' && Buf[I] != '\r' && Buf[I] != '\0') 1247 ++I; 1248 1249 if (Buf[I] == '\n' || Buf[I] == '\r') { 1250 // If this is \r\n, skip both characters. 1251 if (Buf[I] == '\r' && Buf[I+1] == '\n') 1252 ++I; 1253 ++I; 1254 LineOffsets.push_back(I); 1255 } else { 1256 // Otherwise, this is a NUL. If end of file, exit. 1257 if (Buf+I == End) break; 1258 ++I; 1259 } 1260 } 1261 1262 // Copy the offsets into the FileInfo structure. 1263 FI->NumLines = LineOffsets.size(); 1264 FI->SourceLineCache = Alloc.Allocate<unsigned>(LineOffsets.size()); 1265 std::copy(LineOffsets.begin(), LineOffsets.end(), FI->SourceLineCache); 1266 } 1267 1268 /// getLineNumber - Given a SourceLocation, return the spelling line number 1269 /// for the position indicated. This requires building and caching a table of 1270 /// line offsets for the MemoryBuffer, so this is not cheap: use only when 1271 /// about to emit a diagnostic. 1272 unsigned SourceManager::getLineNumber(FileID FID, unsigned FilePos, 1273 bool *Invalid) const { 1274 if (FID.isInvalid()) { 1275 if (Invalid) 1276 *Invalid = true; 1277 return 1; 1278 } 1279 1280 ContentCache *Content; 1281 if (LastLineNoFileIDQuery == FID) 1282 Content = LastLineNoContentCache; 1283 else { 1284 bool MyInvalid = false; 1285 const SLocEntry &Entry = getSLocEntry(FID, &MyInvalid); 1286 if (MyInvalid || !Entry.isFile()) { 1287 if (Invalid) 1288 *Invalid = true; 1289 return 1; 1290 } 1291 1292 Content = const_cast<ContentCache*>(Entry.getFile().getContentCache()); 1293 } 1294 1295 // If this is the first use of line information for this buffer, compute the 1296 /// SourceLineCache for it on demand. 1297 if (!Content->SourceLineCache) { 1298 bool MyInvalid = false; 1299 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid); 1300 if (Invalid) 1301 *Invalid = MyInvalid; 1302 if (MyInvalid) 1303 return 1; 1304 } else if (Invalid) 1305 *Invalid = false; 1306 1307 // Okay, we know we have a line number table. Do a binary search to find the 1308 // line number that this character position lands on. 1309 unsigned *SourceLineCache = Content->SourceLineCache; 1310 unsigned *SourceLineCacheStart = SourceLineCache; 1311 unsigned *SourceLineCacheEnd = SourceLineCache + Content->NumLines; 1312 1313 unsigned QueriedFilePos = FilePos+1; 1314 1315 // FIXME: I would like to be convinced that this code is worth being as 1316 // complicated as it is, binary search isn't that slow. 1317 // 1318 // If it is worth being optimized, then in my opinion it could be more 1319 // performant, simpler, and more obviously correct by just "galloping" outward 1320 // from the queried file position. In fact, this could be incorporated into a 1321 // generic algorithm such as lower_bound_with_hint. 1322 // 1323 // If someone gives me a test case where this matters, and I will do it! - DWD 1324 1325 // If the previous query was to the same file, we know both the file pos from 1326 // that query and the line number returned. This allows us to narrow the 1327 // search space from the entire file to something near the match. 1328 if (LastLineNoFileIDQuery == FID) { 1329 if (QueriedFilePos >= LastLineNoFilePos) { 1330 // FIXME: Potential overflow? 1331 SourceLineCache = SourceLineCache+LastLineNoResult-1; 1332 1333 // The query is likely to be nearby the previous one. Here we check to 1334 // see if it is within 5, 10 or 20 lines. It can be far away in cases 1335 // where big comment blocks and vertical whitespace eat up lines but 1336 // contribute no tokens. 1337 if (SourceLineCache+5 < SourceLineCacheEnd) { 1338 if (SourceLineCache[5] > QueriedFilePos) 1339 SourceLineCacheEnd = SourceLineCache+5; 1340 else if (SourceLineCache+10 < SourceLineCacheEnd) { 1341 if (SourceLineCache[10] > QueriedFilePos) 1342 SourceLineCacheEnd = SourceLineCache+10; 1343 else if (SourceLineCache+20 < SourceLineCacheEnd) { 1344 if (SourceLineCache[20] > QueriedFilePos) 1345 SourceLineCacheEnd = SourceLineCache+20; 1346 } 1347 } 1348 } 1349 } else { 1350 if (LastLineNoResult < Content->NumLines) 1351 SourceLineCacheEnd = SourceLineCache+LastLineNoResult+1; 1352 } 1353 } 1354 1355 unsigned *Pos 1356 = std::lower_bound(SourceLineCache, SourceLineCacheEnd, QueriedFilePos); 1357 unsigned LineNo = Pos-SourceLineCacheStart; 1358 1359 LastLineNoFileIDQuery = FID; 1360 LastLineNoContentCache = Content; 1361 LastLineNoFilePos = QueriedFilePos; 1362 LastLineNoResult = LineNo; 1363 return LineNo; 1364 } 1365 1366 unsigned SourceManager::getSpellingLineNumber(SourceLocation Loc, 1367 bool *Invalid) const { 1368 if (isInvalid(Loc, Invalid)) return 0; 1369 std::pair<FileID, unsigned> LocInfo = getDecomposedSpellingLoc(Loc); 1370 return getLineNumber(LocInfo.first, LocInfo.second); 1371 } 1372 unsigned SourceManager::getExpansionLineNumber(SourceLocation Loc, 1373 bool *Invalid) const { 1374 if (isInvalid(Loc, Invalid)) return 0; 1375 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1376 return getLineNumber(LocInfo.first, LocInfo.second); 1377 } 1378 unsigned SourceManager::getPresumedLineNumber(SourceLocation Loc, 1379 bool *Invalid) const { 1380 PresumedLoc PLoc = getPresumedLoc(Loc); 1381 if (isInvalid(PLoc, Invalid)) return 0; 1382 return PLoc.getLine(); 1383 } 1384 1385 /// getFileCharacteristic - return the file characteristic of the specified 1386 /// source location, indicating whether this is a normal file, a system 1387 /// header, or an "implicit extern C" system header. 1388 /// 1389 /// This state can be modified with flags on GNU linemarker directives like: 1390 /// # 4 "foo.h" 3 1391 /// which changes all source locations in the current file after that to be 1392 /// considered to be from a system header. 1393 SrcMgr::CharacteristicKind 1394 SourceManager::getFileCharacteristic(SourceLocation Loc) const { 1395 assert(Loc.isValid() && "Can't get file characteristic of invalid loc!"); 1396 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1397 bool Invalid = false; 1398 const SLocEntry &SEntry = getSLocEntry(LocInfo.first, &Invalid); 1399 if (Invalid || !SEntry.isFile()) 1400 return C_User; 1401 1402 const SrcMgr::FileInfo &FI = SEntry.getFile(); 1403 1404 // If there are no #line directives in this file, just return the whole-file 1405 // state. 1406 if (!FI.hasLineDirectives()) 1407 return FI.getFileCharacteristic(); 1408 1409 assert(LineTable && "Can't have linetable entries without a LineTable!"); 1410 // See if there is a #line directive before the location. 1411 const LineEntry *Entry = 1412 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second); 1413 1414 // If this is before the first line marker, use the file characteristic. 1415 if (!Entry) 1416 return FI.getFileCharacteristic(); 1417 1418 return Entry->FileKind; 1419 } 1420 1421 /// Return the filename or buffer identifier of the buffer the location is in. 1422 /// Note that this name does not respect \#line directives. Use getPresumedLoc 1423 /// for normal clients. 1424 StringRef SourceManager::getBufferName(SourceLocation Loc, 1425 bool *Invalid) const { 1426 if (isInvalid(Loc, Invalid)) return "<invalid loc>"; 1427 1428 return getBuffer(getFileID(Loc), Invalid)->getBufferIdentifier(); 1429 } 1430 1431 /// getPresumedLoc - This method returns the "presumed" location of a 1432 /// SourceLocation specifies. A "presumed location" can be modified by \#line 1433 /// or GNU line marker directives. This provides a view on the data that a 1434 /// user should see in diagnostics, for example. 1435 /// 1436 /// Note that a presumed location is always given as the expansion point of an 1437 /// expansion location, not at the spelling location. 1438 PresumedLoc SourceManager::getPresumedLoc(SourceLocation Loc, 1439 bool UseLineDirectives) const { 1440 if (Loc.isInvalid()) return PresumedLoc(); 1441 1442 // Presumed locations are always for expansion points. 1443 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1444 1445 bool Invalid = false; 1446 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 1447 if (Invalid || !Entry.isFile()) 1448 return PresumedLoc(); 1449 1450 const SrcMgr::FileInfo &FI = Entry.getFile(); 1451 const SrcMgr::ContentCache *C = FI.getContentCache(); 1452 1453 // To get the source name, first consult the FileEntry (if one exists) 1454 // before the MemBuffer as this will avoid unnecessarily paging in the 1455 // MemBuffer. 1456 FileID FID = LocInfo.first; 1457 StringRef Filename; 1458 if (C->OrigEntry) 1459 Filename = C->OrigEntry->getName(); 1460 else 1461 Filename = C->getBuffer(Diag, getFileManager())->getBufferIdentifier(); 1462 1463 unsigned LineNo = getLineNumber(LocInfo.first, LocInfo.second, &Invalid); 1464 if (Invalid) 1465 return PresumedLoc(); 1466 unsigned ColNo = getColumnNumber(LocInfo.first, LocInfo.second, &Invalid); 1467 if (Invalid) 1468 return PresumedLoc(); 1469 1470 SourceLocation IncludeLoc = FI.getIncludeLoc(); 1471 1472 // If we have #line directives in this file, update and overwrite the physical 1473 // location info if appropriate. 1474 if (UseLineDirectives && FI.hasLineDirectives()) { 1475 assert(LineTable && "Can't have linetable entries without a LineTable!"); 1476 // See if there is a #line directive before this. If so, get it. 1477 if (const LineEntry *Entry = 1478 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) { 1479 // If the LineEntry indicates a filename, use it. 1480 if (Entry->FilenameID != -1) { 1481 Filename = LineTable->getFilename(Entry->FilenameID); 1482 // The contents of files referenced by #line are not in the 1483 // SourceManager 1484 FID = FileID::get(0); 1485 } 1486 1487 // Use the line number specified by the LineEntry. This line number may 1488 // be multiple lines down from the line entry. Add the difference in 1489 // physical line numbers from the query point and the line marker to the 1490 // total. 1491 unsigned MarkerLineNo = getLineNumber(LocInfo.first, Entry->FileOffset); 1492 LineNo = Entry->LineNo + (LineNo-MarkerLineNo-1); 1493 1494 // Note that column numbers are not molested by line markers. 1495 1496 // Handle virtual #include manipulation. 1497 if (Entry->IncludeOffset) { 1498 IncludeLoc = getLocForStartOfFile(LocInfo.first); 1499 IncludeLoc = IncludeLoc.getLocWithOffset(Entry->IncludeOffset); 1500 } 1501 } 1502 } 1503 1504 return PresumedLoc(Filename.data(), FID, LineNo, ColNo, IncludeLoc); 1505 } 1506 1507 /// Returns whether the PresumedLoc for a given SourceLocation is 1508 /// in the main file. 1509 /// 1510 /// This computes the "presumed" location for a SourceLocation, then checks 1511 /// whether it came from a file other than the main file. This is different 1512 /// from isWrittenInMainFile() because it takes line marker directives into 1513 /// account. 1514 bool SourceManager::isInMainFile(SourceLocation Loc) const { 1515 if (Loc.isInvalid()) return false; 1516 1517 // Presumed locations are always for expansion points. 1518 std::pair<FileID, unsigned> LocInfo = getDecomposedExpansionLoc(Loc); 1519 1520 bool Invalid = false; 1521 const SLocEntry &Entry = getSLocEntry(LocInfo.first, &Invalid); 1522 if (Invalid || !Entry.isFile()) 1523 return false; 1524 1525 const SrcMgr::FileInfo &FI = Entry.getFile(); 1526 1527 // Check if there is a line directive for this location. 1528 if (FI.hasLineDirectives()) 1529 if (const LineEntry *Entry = 1530 LineTable->FindNearestLineEntry(LocInfo.first, LocInfo.second)) 1531 if (Entry->IncludeOffset) 1532 return false; 1533 1534 return FI.getIncludeLoc().isInvalid(); 1535 } 1536 1537 /// The size of the SLocEntry that \p FID represents. 1538 unsigned SourceManager::getFileIDSize(FileID FID) const { 1539 bool Invalid = false; 1540 const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1541 if (Invalid) 1542 return 0; 1543 1544 int ID = FID.ID; 1545 unsigned NextOffset; 1546 if ((ID > 0 && unsigned(ID+1) == local_sloc_entry_size())) 1547 NextOffset = getNextLocalOffset(); 1548 else if (ID+1 == -1) 1549 NextOffset = MaxLoadedOffset; 1550 else 1551 NextOffset = getSLocEntry(FileID::get(ID+1)).getOffset(); 1552 1553 return NextOffset - Entry.getOffset() - 1; 1554 } 1555 1556 //===----------------------------------------------------------------------===// 1557 // Other miscellaneous methods. 1558 //===----------------------------------------------------------------------===// 1559 1560 /// Get the source location for the given file:line:col triplet. 1561 /// 1562 /// If the source file is included multiple times, the source location will 1563 /// be based upon an arbitrary inclusion. 1564 SourceLocation SourceManager::translateFileLineCol(const FileEntry *SourceFile, 1565 unsigned Line, 1566 unsigned Col) const { 1567 assert(SourceFile && "Null source file!"); 1568 assert(Line && Col && "Line and column should start from 1!"); 1569 1570 FileID FirstFID = translateFile(SourceFile); 1571 return translateLineCol(FirstFID, Line, Col); 1572 } 1573 1574 /// Get the FileID for the given file. 1575 /// 1576 /// If the source file is included multiple times, the FileID will be the 1577 /// first inclusion. 1578 FileID SourceManager::translateFile(const FileEntry *SourceFile) const { 1579 assert(SourceFile && "Null source file!"); 1580 1581 // First, check the main file ID, since it is common to look for a 1582 // location in the main file. 1583 if (MainFileID.isValid()) { 1584 bool Invalid = false; 1585 const SLocEntry &MainSLoc = getSLocEntry(MainFileID, &Invalid); 1586 if (Invalid) 1587 return FileID(); 1588 1589 if (MainSLoc.isFile()) { 1590 const ContentCache *MainContentCache = 1591 MainSLoc.getFile().getContentCache(); 1592 if (MainContentCache && MainContentCache->OrigEntry == SourceFile) 1593 return MainFileID; 1594 } 1595 } 1596 1597 // The location we're looking for isn't in the main file; look 1598 // through all of the local source locations. 1599 for (unsigned I = 0, N = local_sloc_entry_size(); I != N; ++I) { 1600 bool Invalid = false; 1601 const SLocEntry &SLoc = getLocalSLocEntry(I, &Invalid); 1602 if (Invalid) 1603 return FileID(); 1604 1605 if (SLoc.isFile() && SLoc.getFile().getContentCache() && 1606 SLoc.getFile().getContentCache()->OrigEntry == SourceFile) 1607 return FileID::get(I); 1608 } 1609 1610 // If that still didn't help, try the modules. 1611 for (unsigned I = 0, N = loaded_sloc_entry_size(); I != N; ++I) { 1612 const SLocEntry &SLoc = getLoadedSLocEntry(I); 1613 if (SLoc.isFile() && SLoc.getFile().getContentCache() && 1614 SLoc.getFile().getContentCache()->OrigEntry == SourceFile) 1615 return FileID::get(-int(I) - 2); 1616 } 1617 1618 return FileID(); 1619 } 1620 1621 /// Get the source location in \arg FID for the given line:col. 1622 /// Returns null location if \arg FID is not a file SLocEntry. 1623 SourceLocation SourceManager::translateLineCol(FileID FID, 1624 unsigned Line, 1625 unsigned Col) const { 1626 // Lines are used as a one-based index into a zero-based array. This assert 1627 // checks for possible buffer underruns. 1628 assert(Line && Col && "Line and column should start from 1!"); 1629 1630 if (FID.isInvalid()) 1631 return SourceLocation(); 1632 1633 bool Invalid = false; 1634 const SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1635 if (Invalid) 1636 return SourceLocation(); 1637 1638 if (!Entry.isFile()) 1639 return SourceLocation(); 1640 1641 SourceLocation FileLoc = SourceLocation::getFileLoc(Entry.getOffset()); 1642 1643 if (Line == 1 && Col == 1) 1644 return FileLoc; 1645 1646 ContentCache *Content 1647 = const_cast<ContentCache *>(Entry.getFile().getContentCache()); 1648 if (!Content) 1649 return SourceLocation(); 1650 1651 // If this is the first use of line information for this buffer, compute the 1652 // SourceLineCache for it on demand. 1653 if (!Content->SourceLineCache) { 1654 bool MyInvalid = false; 1655 ComputeLineNumbers(Diag, Content, ContentCacheAlloc, *this, MyInvalid); 1656 if (MyInvalid) 1657 return SourceLocation(); 1658 } 1659 1660 if (Line > Content->NumLines) { 1661 unsigned Size = Content->getBuffer(Diag, getFileManager())->getBufferSize(); 1662 if (Size > 0) 1663 --Size; 1664 return FileLoc.getLocWithOffset(Size); 1665 } 1666 1667 const llvm::MemoryBuffer *Buffer = Content->getBuffer(Diag, getFileManager()); 1668 unsigned FilePos = Content->SourceLineCache[Line - 1]; 1669 const char *Buf = Buffer->getBufferStart() + FilePos; 1670 unsigned BufLength = Buffer->getBufferSize() - FilePos; 1671 if (BufLength == 0) 1672 return FileLoc.getLocWithOffset(FilePos); 1673 1674 unsigned i = 0; 1675 1676 // Check that the given column is valid. 1677 while (i < BufLength-1 && i < Col-1 && Buf[i] != '\n' && Buf[i] != '\r') 1678 ++i; 1679 return FileLoc.getLocWithOffset(FilePos + i); 1680 } 1681 1682 /// Compute a map of macro argument chunks to their expanded source 1683 /// location. Chunks that are not part of a macro argument will map to an 1684 /// invalid source location. e.g. if a file contains one macro argument at 1685 /// offset 100 with length 10, this is how the map will be formed: 1686 /// 0 -> SourceLocation() 1687 /// 100 -> Expanded macro arg location 1688 /// 110 -> SourceLocation() 1689 void SourceManager::computeMacroArgsCache(MacroArgsMap &MacroArgsCache, 1690 FileID FID) const { 1691 assert(FID.isValid()); 1692 1693 // Initially no macro argument chunk is present. 1694 MacroArgsCache.insert(std::make_pair(0, SourceLocation())); 1695 1696 int ID = FID.ID; 1697 while (true) { 1698 ++ID; 1699 // Stop if there are no more FileIDs to check. 1700 if (ID > 0) { 1701 if (unsigned(ID) >= local_sloc_entry_size()) 1702 return; 1703 } else if (ID == -1) { 1704 return; 1705 } 1706 1707 bool Invalid = false; 1708 const SrcMgr::SLocEntry &Entry = getSLocEntryByID(ID, &Invalid); 1709 if (Invalid) 1710 return; 1711 if (Entry.isFile()) { 1712 SourceLocation IncludeLoc = Entry.getFile().getIncludeLoc(); 1713 if (IncludeLoc.isInvalid()) 1714 continue; 1715 if (!isInFileID(IncludeLoc, FID)) 1716 return; // No more files/macros that may be "contained" in this file. 1717 1718 // Skip the files/macros of the #include'd file, we only care about macros 1719 // that lexed macro arguments from our file. 1720 if (Entry.getFile().NumCreatedFIDs) 1721 ID += Entry.getFile().NumCreatedFIDs - 1/*because of next ++ID*/; 1722 continue; 1723 } 1724 1725 const ExpansionInfo &ExpInfo = Entry.getExpansion(); 1726 1727 if (ExpInfo.getExpansionLocStart().isFileID()) { 1728 if (!isInFileID(ExpInfo.getExpansionLocStart(), FID)) 1729 return; // No more files/macros that may be "contained" in this file. 1730 } 1731 1732 if (!ExpInfo.isMacroArgExpansion()) 1733 continue; 1734 1735 associateFileChunkWithMacroArgExp(MacroArgsCache, FID, 1736 ExpInfo.getSpellingLoc(), 1737 SourceLocation::getMacroLoc(Entry.getOffset()), 1738 getFileIDSize(FileID::get(ID))); 1739 } 1740 } 1741 1742 void SourceManager::associateFileChunkWithMacroArgExp( 1743 MacroArgsMap &MacroArgsCache, 1744 FileID FID, 1745 SourceLocation SpellLoc, 1746 SourceLocation ExpansionLoc, 1747 unsigned ExpansionLength) const { 1748 if (!SpellLoc.isFileID()) { 1749 unsigned SpellBeginOffs = SpellLoc.getOffset(); 1750 unsigned SpellEndOffs = SpellBeginOffs + ExpansionLength; 1751 1752 // The spelling range for this macro argument expansion can span multiple 1753 // consecutive FileID entries. Go through each entry contained in the 1754 // spelling range and if one is itself a macro argument expansion, recurse 1755 // and associate the file chunk that it represents. 1756 1757 FileID SpellFID; // Current FileID in the spelling range. 1758 unsigned SpellRelativeOffs; 1759 std::tie(SpellFID, SpellRelativeOffs) = getDecomposedLoc(SpellLoc); 1760 while (true) { 1761 const SLocEntry &Entry = getSLocEntry(SpellFID); 1762 unsigned SpellFIDBeginOffs = Entry.getOffset(); 1763 unsigned SpellFIDSize = getFileIDSize(SpellFID); 1764 unsigned SpellFIDEndOffs = SpellFIDBeginOffs + SpellFIDSize; 1765 const ExpansionInfo &Info = Entry.getExpansion(); 1766 if (Info.isMacroArgExpansion()) { 1767 unsigned CurrSpellLength; 1768 if (SpellFIDEndOffs < SpellEndOffs) 1769 CurrSpellLength = SpellFIDSize - SpellRelativeOffs; 1770 else 1771 CurrSpellLength = ExpansionLength; 1772 associateFileChunkWithMacroArgExp(MacroArgsCache, FID, 1773 Info.getSpellingLoc().getLocWithOffset(SpellRelativeOffs), 1774 ExpansionLoc, CurrSpellLength); 1775 } 1776 1777 if (SpellFIDEndOffs >= SpellEndOffs) 1778 return; // we covered all FileID entries in the spelling range. 1779 1780 // Move to the next FileID entry in the spelling range. 1781 unsigned advance = SpellFIDSize - SpellRelativeOffs + 1; 1782 ExpansionLoc = ExpansionLoc.getLocWithOffset(advance); 1783 ExpansionLength -= advance; 1784 ++SpellFID.ID; 1785 SpellRelativeOffs = 0; 1786 } 1787 } 1788 1789 assert(SpellLoc.isFileID()); 1790 1791 unsigned BeginOffs; 1792 if (!isInFileID(SpellLoc, FID, &BeginOffs)) 1793 return; 1794 1795 unsigned EndOffs = BeginOffs + ExpansionLength; 1796 1797 // Add a new chunk for this macro argument. A previous macro argument chunk 1798 // may have been lexed again, so e.g. if the map is 1799 // 0 -> SourceLocation() 1800 // 100 -> Expanded loc #1 1801 // 110 -> SourceLocation() 1802 // and we found a new macro FileID that lexed from offset 105 with length 3, 1803 // the new map will be: 1804 // 0 -> SourceLocation() 1805 // 100 -> Expanded loc #1 1806 // 105 -> Expanded loc #2 1807 // 108 -> Expanded loc #1 1808 // 110 -> SourceLocation() 1809 // 1810 // Since re-lexed macro chunks will always be the same size or less of 1811 // previous chunks, we only need to find where the ending of the new macro 1812 // chunk is mapped to and update the map with new begin/end mappings. 1813 1814 MacroArgsMap::iterator I = MacroArgsCache.upper_bound(EndOffs); 1815 --I; 1816 SourceLocation EndOffsMappedLoc = I->second; 1817 MacroArgsCache[BeginOffs] = ExpansionLoc; 1818 MacroArgsCache[EndOffs] = EndOffsMappedLoc; 1819 } 1820 1821 /// If \arg Loc points inside a function macro argument, the returned 1822 /// location will be the macro location in which the argument was expanded. 1823 /// If a macro argument is used multiple times, the expanded location will 1824 /// be at the first expansion of the argument. 1825 /// e.g. 1826 /// MY_MACRO(foo); 1827 /// ^ 1828 /// Passing a file location pointing at 'foo', will yield a macro location 1829 /// where 'foo' was expanded into. 1830 SourceLocation 1831 SourceManager::getMacroArgExpandedLocation(SourceLocation Loc) const { 1832 if (Loc.isInvalid() || !Loc.isFileID()) 1833 return Loc; 1834 1835 FileID FID; 1836 unsigned Offset; 1837 std::tie(FID, Offset) = getDecomposedLoc(Loc); 1838 if (FID.isInvalid()) 1839 return Loc; 1840 1841 std::unique_ptr<MacroArgsMap> &MacroArgsCache = MacroArgsCacheMap[FID]; 1842 if (!MacroArgsCache) { 1843 MacroArgsCache = std::make_unique<MacroArgsMap>(); 1844 computeMacroArgsCache(*MacroArgsCache, FID); 1845 } 1846 1847 assert(!MacroArgsCache->empty()); 1848 MacroArgsMap::iterator I = MacroArgsCache->upper_bound(Offset); 1849 --I; 1850 1851 unsigned MacroArgBeginOffs = I->first; 1852 SourceLocation MacroArgExpandedLoc = I->second; 1853 if (MacroArgExpandedLoc.isValid()) 1854 return MacroArgExpandedLoc.getLocWithOffset(Offset - MacroArgBeginOffs); 1855 1856 return Loc; 1857 } 1858 1859 std::pair<FileID, unsigned> 1860 SourceManager::getDecomposedIncludedLoc(FileID FID) const { 1861 if (FID.isInvalid()) 1862 return std::make_pair(FileID(), 0); 1863 1864 // Uses IncludedLocMap to retrieve/cache the decomposed loc. 1865 1866 using DecompTy = std::pair<FileID, unsigned>; 1867 auto InsertOp = IncludedLocMap.try_emplace(FID); 1868 DecompTy &DecompLoc = InsertOp.first->second; 1869 if (!InsertOp.second) 1870 return DecompLoc; // already in map. 1871 1872 SourceLocation UpperLoc; 1873 bool Invalid = false; 1874 const SrcMgr::SLocEntry &Entry = getSLocEntry(FID, &Invalid); 1875 if (!Invalid) { 1876 if (Entry.isExpansion()) 1877 UpperLoc = Entry.getExpansion().getExpansionLocStart(); 1878 else 1879 UpperLoc = Entry.getFile().getIncludeLoc(); 1880 } 1881 1882 if (UpperLoc.isValid()) 1883 DecompLoc = getDecomposedLoc(UpperLoc); 1884 1885 return DecompLoc; 1886 } 1887 1888 /// Given a decomposed source location, move it up the include/expansion stack 1889 /// to the parent source location. If this is possible, return the decomposed 1890 /// version of the parent in Loc and return false. If Loc is the top-level 1891 /// entry, return true and don't modify it. 1892 static bool MoveUpIncludeHierarchy(std::pair<FileID, unsigned> &Loc, 1893 const SourceManager &SM) { 1894 std::pair<FileID, unsigned> UpperLoc = SM.getDecomposedIncludedLoc(Loc.first); 1895 if (UpperLoc.first.isInvalid()) 1896 return true; // We reached the top. 1897 1898 Loc = UpperLoc; 1899 return false; 1900 } 1901 1902 /// Return the cache entry for comparing the given file IDs 1903 /// for isBeforeInTranslationUnit. 1904 InBeforeInTUCacheEntry &SourceManager::getInBeforeInTUCache(FileID LFID, 1905 FileID RFID) const { 1906 // This is a magic number for limiting the cache size. It was experimentally 1907 // derived from a small Objective-C project (where the cache filled 1908 // out to ~250 items). We can make it larger if necessary. 1909 enum { MagicCacheSize = 300 }; 1910 IsBeforeInTUCacheKey Key(LFID, RFID); 1911 1912 // If the cache size isn't too large, do a lookup and if necessary default 1913 // construct an entry. We can then return it to the caller for direct 1914 // use. When they update the value, the cache will get automatically 1915 // updated as well. 1916 if (IBTUCache.size() < MagicCacheSize) 1917 return IBTUCache[Key]; 1918 1919 // Otherwise, do a lookup that will not construct a new value. 1920 InBeforeInTUCache::iterator I = IBTUCache.find(Key); 1921 if (I != IBTUCache.end()) 1922 return I->second; 1923 1924 // Fall back to the overflow value. 1925 return IBTUCacheOverflow; 1926 } 1927 1928 /// Determines the order of 2 source locations in the translation unit. 1929 /// 1930 /// \returns true if LHS source location comes before RHS, false otherwise. 1931 bool SourceManager::isBeforeInTranslationUnit(SourceLocation LHS, 1932 SourceLocation RHS) const { 1933 assert(LHS.isValid() && RHS.isValid() && "Passed invalid source location!"); 1934 if (LHS == RHS) 1935 return false; 1936 1937 std::pair<FileID, unsigned> LOffs = getDecomposedLoc(LHS); 1938 std::pair<FileID, unsigned> ROffs = getDecomposedLoc(RHS); 1939 1940 // getDecomposedLoc may have failed to return a valid FileID because, e.g. it 1941 // is a serialized one referring to a file that was removed after we loaded 1942 // the PCH. 1943 if (LOffs.first.isInvalid() || ROffs.first.isInvalid()) 1944 return LOffs.first.isInvalid() && !ROffs.first.isInvalid(); 1945 1946 std::pair<bool, bool> InSameTU = isInTheSameTranslationUnit(LOffs, ROffs); 1947 if (InSameTU.first) 1948 return InSameTU.second; 1949 1950 // If we arrived here, the location is either in a built-ins buffer or 1951 // associated with global inline asm. PR5662 and PR22576 are examples. 1952 1953 StringRef LB = getBuffer(LOffs.first)->getBufferIdentifier(); 1954 StringRef RB = getBuffer(ROffs.first)->getBufferIdentifier(); 1955 bool LIsBuiltins = LB == "<built-in>"; 1956 bool RIsBuiltins = RB == "<built-in>"; 1957 // Sort built-in before non-built-in. 1958 if (LIsBuiltins || RIsBuiltins) { 1959 if (LIsBuiltins != RIsBuiltins) 1960 return LIsBuiltins; 1961 // Both are in built-in buffers, but from different files. We just claim that 1962 // lower IDs come first. 1963 return LOffs.first < ROffs.first; 1964 } 1965 bool LIsAsm = LB == "<inline asm>"; 1966 bool RIsAsm = RB == "<inline asm>"; 1967 // Sort assembler after built-ins, but before the rest. 1968 if (LIsAsm || RIsAsm) { 1969 if (LIsAsm != RIsAsm) 1970 return RIsAsm; 1971 assert(LOffs.first == ROffs.first); 1972 return false; 1973 } 1974 bool LIsScratch = LB == "<scratch space>"; 1975 bool RIsScratch = RB == "<scratch space>"; 1976 // Sort scratch after inline asm, but before the rest. 1977 if (LIsScratch || RIsScratch) { 1978 if (LIsScratch != RIsScratch) 1979 return LIsScratch; 1980 return LOffs.second < ROffs.second; 1981 } 1982 llvm_unreachable("Unsortable locations found"); 1983 } 1984 1985 std::pair<bool, bool> SourceManager::isInTheSameTranslationUnit( 1986 std::pair<FileID, unsigned> &LOffs, 1987 std::pair<FileID, unsigned> &ROffs) const { 1988 // If the source locations are in the same file, just compare offsets. 1989 if (LOffs.first == ROffs.first) 1990 return std::make_pair(true, LOffs.second < ROffs.second); 1991 1992 // If we are comparing a source location with multiple locations in the same 1993 // file, we get a big win by caching the result. 1994 InBeforeInTUCacheEntry &IsBeforeInTUCache = 1995 getInBeforeInTUCache(LOffs.first, ROffs.first); 1996 1997 // If we are comparing a source location with multiple locations in the same 1998 // file, we get a big win by caching the result. 1999 if (IsBeforeInTUCache.isCacheValid(LOffs.first, ROffs.first)) 2000 return std::make_pair( 2001 true, IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second)); 2002 2003 // Okay, we missed in the cache, start updating the cache for this query. 2004 IsBeforeInTUCache.setQueryFIDs(LOffs.first, ROffs.first, 2005 /*isLFIDBeforeRFID=*/LOffs.first.ID < ROffs.first.ID); 2006 2007 // We need to find the common ancestor. The only way of doing this is to 2008 // build the complete include chain for one and then walking up the chain 2009 // of the other looking for a match. 2010 // We use a map from FileID to Offset to store the chain. Easier than writing 2011 // a custom set hash info that only depends on the first part of a pair. 2012 using LocSet = llvm::SmallDenseMap<FileID, unsigned, 16>; 2013 LocSet LChain; 2014 do { 2015 LChain.insert(LOffs); 2016 // We catch the case where LOffs is in a file included by ROffs and 2017 // quit early. The other way round unfortunately remains suboptimal. 2018 } while (LOffs.first != ROffs.first && !MoveUpIncludeHierarchy(LOffs, *this)); 2019 LocSet::iterator I; 2020 while((I = LChain.find(ROffs.first)) == LChain.end()) { 2021 if (MoveUpIncludeHierarchy(ROffs, *this)) 2022 break; // Met at topmost file. 2023 } 2024 if (I != LChain.end()) 2025 LOffs = *I; 2026 2027 // If we exited because we found a nearest common ancestor, compare the 2028 // locations within the common file and cache them. 2029 if (LOffs.first == ROffs.first) { 2030 IsBeforeInTUCache.setCommonLoc(LOffs.first, LOffs.second, ROffs.second); 2031 return std::make_pair( 2032 true, IsBeforeInTUCache.getCachedResult(LOffs.second, ROffs.second)); 2033 } 2034 // Clear the lookup cache, it depends on a common location. 2035 IsBeforeInTUCache.clear(); 2036 return std::make_pair(false, false); 2037 } 2038 2039 void SourceManager::PrintStats() const { 2040 llvm::errs() << "\n*** Source Manager Stats:\n"; 2041 llvm::errs() << FileInfos.size() << " files mapped, " << MemBufferInfos.size() 2042 << " mem buffers mapped.\n"; 2043 llvm::errs() << LocalSLocEntryTable.size() << " local SLocEntry's allocated (" 2044 << llvm::capacity_in_bytes(LocalSLocEntryTable) 2045 << " bytes of capacity), " 2046 << NextLocalOffset << "B of Sloc address space used.\n"; 2047 llvm::errs() << LoadedSLocEntryTable.size() 2048 << " loaded SLocEntries allocated, " 2049 << MaxLoadedOffset - CurrentLoadedOffset 2050 << "B of Sloc address space used.\n"; 2051 2052 unsigned NumLineNumsComputed = 0; 2053 unsigned NumFileBytesMapped = 0; 2054 for (fileinfo_iterator I = fileinfo_begin(), E = fileinfo_end(); I != E; ++I){ 2055 NumLineNumsComputed += I->second->SourceLineCache != nullptr; 2056 NumFileBytesMapped += I->second->getSizeBytesMapped(); 2057 } 2058 unsigned NumMacroArgsComputed = MacroArgsCacheMap.size(); 2059 2060 llvm::errs() << NumFileBytesMapped << " bytes of files mapped, " 2061 << NumLineNumsComputed << " files with line #'s computed, " 2062 << NumMacroArgsComputed << " files with macro args computed.\n"; 2063 llvm::errs() << "FileID scans: " << NumLinearScans << " linear, " 2064 << NumBinaryProbes << " binary.\n"; 2065 } 2066 2067 LLVM_DUMP_METHOD void SourceManager::dump() const { 2068 llvm::raw_ostream &out = llvm::errs(); 2069 2070 auto DumpSLocEntry = [&](int ID, const SrcMgr::SLocEntry &Entry, 2071 llvm::Optional<unsigned> NextStart) { 2072 out << "SLocEntry <FileID " << ID << "> " << (Entry.isFile() ? "file" : "expansion") 2073 << " <SourceLocation " << Entry.getOffset() << ":"; 2074 if (NextStart) 2075 out << *NextStart << ">\n"; 2076 else 2077 out << "???\?>\n"; 2078 if (Entry.isFile()) { 2079 auto &FI = Entry.getFile(); 2080 if (FI.NumCreatedFIDs) 2081 out << " covers <FileID " << ID << ":" << int(ID + FI.NumCreatedFIDs) 2082 << ">\n"; 2083 if (FI.getIncludeLoc().isValid()) 2084 out << " included from " << FI.getIncludeLoc().getOffset() << "\n"; 2085 if (auto *CC = FI.getContentCache()) { 2086 out << " for " << (CC->OrigEntry ? CC->OrigEntry->getName() : "<none>") 2087 << "\n"; 2088 if (CC->BufferOverridden) 2089 out << " contents overridden\n"; 2090 if (CC->ContentsEntry != CC->OrigEntry) { 2091 out << " contents from " 2092 << (CC->ContentsEntry ? CC->ContentsEntry->getName() : "<none>") 2093 << "\n"; 2094 } 2095 } 2096 } else { 2097 auto &EI = Entry.getExpansion(); 2098 out << " spelling from " << EI.getSpellingLoc().getOffset() << "\n"; 2099 out << " macro " << (EI.isMacroArgExpansion() ? "arg" : "body") 2100 << " range <" << EI.getExpansionLocStart().getOffset() << ":" 2101 << EI.getExpansionLocEnd().getOffset() << ">\n"; 2102 } 2103 }; 2104 2105 // Dump local SLocEntries. 2106 for (unsigned ID = 0, NumIDs = LocalSLocEntryTable.size(); ID != NumIDs; ++ID) { 2107 DumpSLocEntry(ID, LocalSLocEntryTable[ID], 2108 ID == NumIDs - 1 ? NextLocalOffset 2109 : LocalSLocEntryTable[ID + 1].getOffset()); 2110 } 2111 // Dump loaded SLocEntries. 2112 llvm::Optional<unsigned> NextStart; 2113 for (unsigned Index = 0; Index != LoadedSLocEntryTable.size(); ++Index) { 2114 int ID = -(int)Index - 2; 2115 if (SLocEntryLoaded[Index]) { 2116 DumpSLocEntry(ID, LoadedSLocEntryTable[Index], NextStart); 2117 NextStart = LoadedSLocEntryTable[Index].getOffset(); 2118 } else { 2119 NextStart = None; 2120 } 2121 } 2122 } 2123 2124 ExternalSLocEntrySource::~ExternalSLocEntrySource() = default; 2125 2126 /// Return the amount of memory used by memory buffers, breaking down 2127 /// by heap-backed versus mmap'ed memory. 2128 SourceManager::MemoryBufferSizes SourceManager::getMemoryBufferSizes() const { 2129 size_t malloc_bytes = 0; 2130 size_t mmap_bytes = 0; 2131 2132 for (unsigned i = 0, e = MemBufferInfos.size(); i != e; ++i) 2133 if (size_t sized_mapped = MemBufferInfos[i]->getSizeBytesMapped()) 2134 switch (MemBufferInfos[i]->getMemoryBufferKind()) { 2135 case llvm::MemoryBuffer::MemoryBuffer_MMap: 2136 mmap_bytes += sized_mapped; 2137 break; 2138 case llvm::MemoryBuffer::MemoryBuffer_Malloc: 2139 malloc_bytes += sized_mapped; 2140 break; 2141 } 2142 2143 return MemoryBufferSizes(malloc_bytes, mmap_bytes); 2144 } 2145 2146 size_t SourceManager::getDataStructureSizes() const { 2147 size_t size = llvm::capacity_in_bytes(MemBufferInfos) 2148 + llvm::capacity_in_bytes(LocalSLocEntryTable) 2149 + llvm::capacity_in_bytes(LoadedSLocEntryTable) 2150 + llvm::capacity_in_bytes(SLocEntryLoaded) 2151 + llvm::capacity_in_bytes(FileInfos); 2152 2153 if (OverriddenFilesInfo) 2154 size += llvm::capacity_in_bytes(OverriddenFilesInfo->OverriddenFiles); 2155 2156 return size; 2157 } 2158 2159 SourceManagerForFile::SourceManagerForFile(StringRef FileName, 2160 StringRef Content) { 2161 // This is referenced by `FileMgr` and will be released by `FileMgr` when it 2162 // is deleted. 2163 IntrusiveRefCntPtr<llvm::vfs::InMemoryFileSystem> InMemoryFileSystem( 2164 new llvm::vfs::InMemoryFileSystem); 2165 InMemoryFileSystem->addFile( 2166 FileName, 0, 2167 llvm::MemoryBuffer::getMemBuffer(Content, FileName, 2168 /*RequiresNullTerminator=*/false)); 2169 // This is passed to `SM` as reference, so the pointer has to be referenced 2170 // in `Environment` so that `FileMgr` can out-live this function scope. 2171 FileMgr = 2172 std::make_unique<FileManager>(FileSystemOptions(), InMemoryFileSystem); 2173 // This is passed to `SM` as reference, so the pointer has to be referenced 2174 // by `Environment` due to the same reason above. 2175 Diagnostics = std::make_unique<DiagnosticsEngine>( 2176 IntrusiveRefCntPtr<DiagnosticIDs>(new DiagnosticIDs), 2177 new DiagnosticOptions); 2178 SourceMgr = std::make_unique<SourceManager>(*Diagnostics, *FileMgr); 2179 FileID ID = SourceMgr->createFileID(*FileMgr->getFile(FileName), 2180 SourceLocation(), clang::SrcMgr::C_User); 2181 assert(ID.isValid()); 2182 SourceMgr->setMainFileID(ID); 2183 } 2184